## Vibhav Gogate## Iterative Join Graph Propagation## PapersRobert Mattescu, Kalev Kask, Vibhav Gogate and Rina Dechter, “Join-Graph Propagation Algorithms/,” Journal of Artificial Intelligence Research, 2010. [ **PDF**].
## Help
ijgp [options] < uai08file > < uai08evidfile > The uai08file and the uai08evidfile should be in UAI 2008 format. The description of this format is available here. The output will be stored in < uai08file > .MAR in the working directory. Note that after each i-bound, the results are written to the output file. [http:www.cs.huji.ac.il
--i-bound [int] (default value is 0. The starting i-bound used by IJGP.) --num-iterations [int] (default value is 10. The number of iterations for which IJGP is run.) --order [min-fill|min-degree|lex] (default value is min-fill. The ordering heuristic used to construct the join graph.) --num-tries [int] (The orderings are randomized. default value is 1. If set to a value greater than 1, then we run the ordering heuristic a number of times and select the one that yields the smallest treewidth.)
## Exampleijgp –order min-degree –i-bound 3 –num-iterations 100 BN_69.uai BN_69.uai.evid This will run ijgp on the specified files. The join-graphs will be constructed using the min-degree ordering. The starting i-bound is 3. Note that IJGP is run as an anytime scheme. In this case, we will start with an i-bound of 3, run IJGP until convergence or 100 iterations whichever is earlier. Then we increase the i-bound by one and reconstruct the join graph, iteratively repeating the process until we run out of memory or time. |